3D CNN in Keras - Action Recognition

# The code for 3D CNN for Action Recognition
# Please refer to the youtube video for this lesson

3D CNN-Action Recognition Part-1

3D CNN-Action Recognition Part-2

from keras.preprocessing.image import ImageDataGenerator
from keras.models import Sequential
from keras.layers.core import Dense, Dropout, Activation, Flatten
from keras.layers.convolutional import Convolution3D, MaxPooling3D

from keras.optimizers import SGD, RMSprop
from keras.utils import np_utils, generic_utils

import theano
import os
import matplotlib
import matplotlib.pyplot as plt
import numpy as np
import cv2
from sklearn.cross_validation import train_test_split
from sklearn import cross_validation
from sklearn import preprocessing


# image specification
img_rows,img_cols,img_depth=16,16,15


# Training data

X_tr=[]           # variable to store entire dataset

#Reading boxing action class

listing = os.listdir('kth dataset/boxing')

for vid in listing:
    vid = 'kth dataset/boxing/'+vid
    frames = []
    cap = cv2.VideoCapture(vid)
    fps = cap.get(5)
    print "Frames per second using video.get(cv2.cv.CV_CAP_PROP_FPS): {0}".format(fps)
 

    for k in xrange(15):
        ret, frame = cap.read()
        frame=cv2.resize(frame,(img_rows,img_cols),interpolation=cv2.INTER_AREA)
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
         frames.append(gray)

        #plt.imshow(gray, cmap = plt.get_cmap('gray'))
        #plt.xticks([]), plt.yticks([])  # to hide tick values on X and Y axis
        #plt.show()
        #cv2.imshow('frame',gray)

        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    cap.release()
    cv2.destroyAllWindows()

    input=np.array(frames)

    print input.shape
    ipt=np.rollaxis(np.rollaxis(input,2,0),2,0)
    print ipt.shape

    X_tr.append(ipt)

#Reading hand clapping action class

listing2 = os.listdir('kth dataset/handclapping')

for vid2 in listing2:
    vid2 = 'kth dataset/handclapping/'+vid2
    frames = []
    cap = cv2.VideoCapture(vid2)
    fps = cap.get(5)
    print "Frames per second using video.get(cv2.cv.CV_CAP_PROP_FPS): {0}".format(fps)

    for k in xrange(15):
        ret, frame = cap.read()
        frame=cv2.resize(frame,(img_rows,img_cols),interpolation=cv2.INTER_AREA)
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        frames.append(gray)

        #plt.imshow(gray, cmap = plt.get_cmap('gray'))
        #plt.xticks([]), plt.yticks([])  # to hide tick values on X and Y axis
        #plt.show()
        #cv2.imshow('frame',gray)

        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    cap.release()
    cv2.destroyAllWindows()
    input=np.array(frames)

    print input.shape
    ipt=np.rollaxis(np.rollaxis(input,2,0),2,0)
    print ipt.shape

    X_tr.append(ipt)

#Reading hand waving action class

listing3 = os.listdir('kth dataset/handwaving')

for vid3 in listing3:
    vid3 = 'kth dataset/handwaving/'+vid3
    frames = []
    cap = cv2.VideoCapture(vid3)
    fps = cap.get(5)
    print "Frames per second using video.get(cv2.cv.CV_CAP_PROP_FPS): {0}".format(fps)

    for k in xrange(15):
        ret, frame = cap.read()
        frame=cv2.resize(frame,(img_rows,img_cols),interpolation=cv2.INTER_AREA)
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        frames.append(gray)

        #plt.imshow(gray, cmap = plt.get_cmap('gray'))
        #plt.xticks([]), plt.yticks([])  # to hide tick values on X and Y axis
        #plt.show()
        #cv2.imshow('frame',gray)

        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    cap.release()
    cv2.destroyAllWindows()
    input=np.array(frames)

    print input.shape
    ipt=np.rollaxis(np.rollaxis(input,2,0),2,0)
    print ipt.shape

    X_tr.append(ipt)

#Reading jogging action class

listing4 = os.listdir('kth dataset/jogging')

for vid4 in listing4:
    vid4 = 'kth dataset/jogging/'+vid4
    frames = []
    cap = cv2.VideoCapture(vid4)
    fps = cap.get(5)
    print "Frames per second using video.get(cv2.cv.CV_CAP_PROP_FPS): {0}".format(fps)

    for k in xrange(15):
        ret, frame = cap.read()
        frame=cv2.resize(frame,(img_rows,img_cols),interpolation=cv2.INTER_AREA)
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        frames.append(gray)

        #plt.imshow(gray, cmap = plt.get_cmap('gray'))
        #plt.xticks([]), plt.yticks([])  # to hide tick values on X and Y axis
        #plt.show()
        #cv2.imshow('frame',gray)

        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    cap.release()
    cv2.destroyAllWindows()
    input=np.array(frames)

    print input.shape
    ipt=np.rollaxis(np.rollaxis(input,2,0),2,0)
    print ipt.shape

    X_tr.append(ipt)

#Reading running action class
 
listing5 = os.listdir('kth dataset/running')

for vid5 in listing5:
    vid5 = 'kth dataset/running/'+vid5
    frames = []
    cap = cv2.VideoCapture(vid5)
    fps = cap.get(5)
    print "Frames per second using video.get(cv2.cv.CV_CAP_PROP_FPS): {0}".format(fps)

    for k in xrange(15):
        ret, frame = cap.read()
        frame=cv2.resize(frame,(img_rows,img_cols),interpolation=cv2.INTER_AREA)
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        frames.append(gray)

        #plt.imshow(gray, cmap = plt.get_cmap('gray'))
        #plt.xticks([]), plt.yticks([])  # to hide tick values on X and Y axis
        #plt.show()
        #cv2.imshow('frame',gray)

        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    cap.release()
    cv2.destroyAllWindows()
    input=np.array(frames)

    print input.shape
    ipt=np.rollaxis(np.rollaxis(input,2,0),2,0)
    print ipt.shape

    X_tr.append(ipt)
 

#Reading walking action class 

listing6 = os.listdir('kth dataset/walking')

for vid6 in listing6:
    vid6 = 'kth dataset/walking/'+vid6
    frames = []
    cap = cv2.VideoCapture(vid6)
    fps = cap.get(5)
    print "Frames per second using video.get(cv2.cv.CV_CAP_PROP_FPS): {0}".format(fps)

    for k in xrange(15):
        ret, frame = cap.read()
        frame=cv2.resize(frame,(img_rows,img_cols),interpolation=cv2.INTER_AREA)
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        frames.append(gray)

        #plt.imshow(gray, cmap = plt.get_cmap('gray'))
        #plt.xticks([]), plt.yticks([])  # to hide tick values on X and Y axis
        #plt.show()
        #cv2.imshow('frame',gray)

        if cv2.waitKey(1) & 0xFF == ord('q'):
            break
    cap.release()
    cv2.destroyAllWindows()
    input=np.array(frames)

    print input.shape
    ipt=np.rollaxis(np.rollaxis(input,2,0),2,0)
    print ipt.shape

    X_tr.append(ipt)



X_tr_array = np.array(X_tr)   # convert the frames read into array

num_samples = len(X_tr_array)
print num_samples

#Assign Label to each class

label=np.ones((num_samples,),dtype = int)
label[0:100]= 0
label[100:199] = 1
label[199:299] = 2
label[299:399] = 3
label[399:499]= 4
label[499:] = 5


train_data = [X_tr_array,label]

(X_train, y_train) = (train_data[0],train_data[1])
print('X_Train shape:', X_train.shape)

train_set = np.zeros((num_samples, 1, img_rows,img_cols,img_depth))

for h in xrange(num_samples):
    train_set[h][0][:][:][:]=X_train[h,:,:,:]
 

patch_size = 15    # img_depth or number of frames used for each video

print(train_set.shape, 'train samples')

# CNN Training parameters

batch_size = 2
nb_classes = 6
nb_epoch =50

# convert class vectors to binary class matrices
Y_train = np_utils.to_categorical(y_train, nb_classes)


# number of convolutional filters to use at each layer
nb_filters = [32, 32]

# level of pooling to perform at each layer (POOL x POOL)
nb_pool = [3, 3]

# level of convolution to perform at each layer (CONV x CONV)
nb_conv = [5,5]

# Pre-processing

train_set = train_set.astype('float32')

train_set -= np.mean(train_set)

train_set /=np.max(train_set)



# Define model

model = Sequential()
model.add(Convolution3D(nb_filters[0],nb_depth=nb_conv[0], nb_row=nb_conv[0], nb_col=nb_conv[0], input_shape=(1, img_rows, img_cols, patch_size), activation='relu'))

model.add(MaxPooling3D(pool_size=(nb_pool[0], nb_pool[0], nb_pool[0])))

model.add(Dropout(0.5))

model.add(Flatten())

model.add(Dense(128, init='normal', activation='relu'))

model.add(Dropout(0.5))

model.add(Dense(nb_classes,init='normal'))

model.add(Activation('softmax'))

model.compile(loss='categorical_crossentropy', optimizer='RMSprop')

 
# Split the data

X_train_new, X_val_new, y_train_new,y_val_new =  train_test_split(train_set, Y_train, test_size=0.2, random_state=4)


# Train the model

hist = model.fit(X_train_new, y_train_new, validation_data=(X_val_new,y_val_new),
          batch_size=batch_size,nb_epoch = nb_epoch,show_accuracy=True,shuffle=True)


#hist = model.fit(train_set, Y_train, batch_size=batch_size,
#         nb_epoch=nb_epoch,validation_split=0.2, show_accuracy=True,
#           shuffle=True)


  # Evaluate the model
score = model.evaluate(X_val_new, y_val_new, batch_size=batch_size, show_accuracy=True)
print('Test score:', score[0])
print('Test accuracy:', score[1])



# Plot the results
train_loss=hist.history['loss']
val_loss=hist.history['val_loss']
train_acc=hist.history['acc']
val_acc=hist.history['val_acc']
xc=range(100)

plt.figure(1,figsize=(7,5))
plt.plot(xc,train_loss)
plt.plot(xc,val_loss)
plt.xlabel('num of Epochs')
plt.ylabel('loss')
plt.title('train_loss vs val_loss')
plt.grid(True)
plt.legend(['train','val'])
print plt.style.available # use bmh, classic,ggplot for big pictures
plt.style.use(['classic'])

plt.figure(2,figsize=(7,5))
plt.plot(xc,train_acc)
plt.plot(xc,val_acc)
plt.xlabel('num of Epochs')
plt.ylabel('accuracy')
plt.title('train_acc vs val_acc')
plt.grid(True)
plt.legend(['train','val'],loc=4)
#print plt.style.available # use bmh, classic,ggplot for big pictures

plt.style.use(['classic'])

原文地址：http：//learnandshare645.blogspot.in/2016/06/3d-cnn-in-keras-action-recognition.html